theo_em-optim-repro Qwen3-8B bad-medical Muon bs32 2ep specreg

A Qwen3-8B fine-tune produced by the Misalignment-Empirics research group (LASR Labs Summer 2026 cohort) as part of an optimizer sweep on a 'bad medical' dataset, studying whether and how different training configurations induce broad emergent misalignment in a large language model. This artifact is a research checkpoint, not a general-purpose assistant. Source: supabase.